July 2, 2026 · Series manifesto
One Developer vs. the Consumer Internet
I am rebuilding the product experiences behind the world's biggest consumer apps, one functional vertical slice at a time, to test what a single developer working with AI can now ship.
The goal is deliberately unreasonable: use HAAM to rebuild the recognisable product experience of every major consumer app and website, then publish the evidence of what one developer could and could not reproduce.
This is not a portfolio of clone tutorials. It is a public benchmark for the new economics of software creation. AI can now generate code, interfaces, content, tests, documentation, and data transformations at a speed that would have required a larger team only a few years ago. The useful question is no longer whether a single person can make a screen that resembles a famous app. The useful question is how much of the real product loop one person can make coherent, dependable, accessible, fast, and honest.
The series will also show where the interface is the easy part. A polished demo does not reproduce a marketplace, a social graph, music rights, fraud operations, logistics, moderation, global support, or years of behavioral data. Those missing systems are part of the product, and each article will name them directly.
Why these targets
Popularity gets attention. Product range creates the test.
The target universe starts with products that dominate global web traffic and app downloads, but the build order is strategic rather than numerical. Google, YouTube, Facebook, Instagram, and ChatGPT sit near the top of global website traffic, while ChatGPT, Instagram, and TikTok led global app downloads in 2025. Rebuilding them in traffic order would produce several social and infrastructure-heavy demos before the series had established a useful method.
The opening targets therefore favor products with a visible end-to-end journey. They make it possible to test whether one developer can connect discovery, decision support, transactions, account states, accessibility, AI assistance, and responsive design into one credible experience.
The selection model
Recognisable in one screen
The target should be familiar enough that people can judge the result without a long explanation.
A complete user loop
The rebuild needs discovery, decision, action, feedback, and recovery instead of a static homepage imitation.
Several kinds of product work
Strong targets combine interaction design, frontend engineering, data, AI, accessibility, trust, performance, and localization.
A useful vertical slice
The experience should remain meaningful with synthetic data and without pretending that a global marketplace or social graph already exists.
Room to improve the original pattern
The point is not pixel matching. Each rebuild should test a clearer, calmer, more accessible, or more transparent product direction.
Initial target map
Thirty product systems worth taking apart
Wave 1
Full product loops
The best opening targets. Each can become a credible working product without first recreating a global social graph.
Airbnb
Travel marketplace, maps, trust, checkout, trips, hosting
Duolingo
Learning loops, speech, personalization, streaks, content generation
Communities, ranking, moderation, identity, threaded discussion
Strava
Activity data, maps, goals, social proof, subscriptions
Visual discovery, recommendations, boards, commerce intent
Booking.com
Dense search, comparison, urgency, pricing, localization
Uber Eats
Local discovery, menus, cart, delivery states, support
Letterboxd
Catalog, reviews, lists, taste graphs, community identity
Wave 2
Media and social systems
Highly visible interfaces where the demo must clearly separate the product experience from the network effects and media rights that power the real service.
Feed, stories, creation, messaging, recommendations
TikTok
Vertical video, editing, ranking, safety, creator tools
YouTube
Search, playback, channels, comments, creator analytics
Spotify
Catalog browsing, playlists, recommendations, playback states
Netflix
Profiles, discovery, playback, household context, retention
Twitch
Live video, chat, subscriptions, moderation, communities
Discord
Servers, channels, voice states, roles, community management
Private messaging, groups, media, calls, business entry points
Tinder
Profiles, matching, safety, messaging, paid visibility
Identity, groups, marketplace, events, feed, moderation
Wave 3
Infrastructure-shaped products
The interface can be rebuilt, but the article must expose how much value comes from data, logistics, regulation, capital, and global operations behind the screen.
Google Search
Query understanding, ranking, answer interfaces, advertising
Google Maps
Geospatial search, routing, places, reviews, live context
Amazon
Catalog, search, recommendations, checkout, logistics visibility
Uber
Realtime location, matching, pricing, safety, trip states
ChatGPT
Conversation, tools, memory, files, multimodal interaction
Canva
Editor architecture, templates, collaboration, generative media
Professional identity, feed, jobs, messaging, recruiting
X
Realtime public conversation, ranking, communities, moderation
Wikipedia
Knowledge navigation, editing, citations, governance
Steam
Store, library, community, updates, reviews, marketplace
Etsy
Search, seller identity, customization, checkout, trust
Temu
Discovery loops, promotion mechanics, cart, logistics states
Target 01
Airbnb
Airbnb is the strongest first test because a useful version can contain a complete product loop: search, maps, filters, listings, trust, availability, checkout, saved trips, messaging, and host tools. It is instantly recognisable, but a convincing demo does not require a live social feed or licensed media catalog.
Every rebuild gets the same scoreboard
Rebuild the product pattern, not the brand identity
Each project will use an original name, visual system, codebase, and dataset. The famous product name appears in the editorial framing so readers understand the reference point, not as a claim of authorization or affiliation. No private source code, copied logo, scraped personal data, or confusing store listing belongs in this experiment.
That line matters creatively as well as legally. A literal copy only proves that AI can imitate a screenshot. A distinct reimplementation can reveal the underlying product grammar and ask whether the experience could be better.